ANN, Fuzzy, ANFIS, MPPT, Photovoltaic System, Buck-boost converter

Photovoltaic (PV) systems have shown growth in the world's electrical matrix. However, the non-linear nature of PV array and their strong dependence on ambient conditions decrease the maximum power they can produce and, consequently, reduce their performance and commercial attractiveness. Maximum Power Point Tracking (MPPT) techniques have been studied over the years to minimize these problems. Among the various control techniques used for spot tracking and maximum power, those that use intelligent algorithms to control the switching of DC-DC converters have shown a high potential for use. Therefore, the present work proposes to develop MPPT techniques based on Artificial Neural Network (ANN), fuzzy and Adaptive Neuro-Fuzzy Inference System (ANFIS) intelligent algorithms, to be applied to PV systems that have the buck-boost as a CC-CC converter. Three proposed architectures were developed for each algorithm. They were compared with each other and with the classic Perturb and Observe (P&O) algorithm. The proposals are distinguished by the input variables used, namely: irradiance and ambient temperature, for Proposal 1, with purely environmental parameters as input variables; irradiance and instantaneous output power of the PV array, for Proposal 2, with input variables that mixed environmental and electrical parameters; and instantaneous and previous instantaneous output power of the PV array, for Proposal 3, with purely electrical parameters as input variables. To assist in the study of the performance of the intelligent algorithms, two scenarios of PV systems, composed of PV array, buck-boost converter, MPPT and load, were modeled, identified Scenario 1 and Scenario 2. The scenarios were differentiated by the total power of the system. At the end of the analyses, it was noticed that the intelligent algorithms had a high tracking speed and were more stable than the P&O algorithms. The PV systems controlled by the intelligent algorithms of Proposal 1 showed the highest efficiency in reaching the maximum power point. The ANFIS and ANN algorithms were more prominent. In power generation, ANN recovered up to 12.05% of the energy lost when using P&O. In the Proposal 2 study, the PV systems also performed well, but lower than the Proposal 1 algorithms. The highest power generated was also achieved by the ANN. It generated 12.01% more power than the P&O. In Proposal 3, the intelligent algorithms had their efficiency compromised because the overlapping of some database values. Anyway, under Random condition, the intelligent algorithms still proved to be superior to P&O in tracking the maximum power point, recovering 8.27% of the generated power. Therefore, intelligent algorithms, especially ANN and ANFIS, have shown the relevance of their use in photovoltaic applications, especially in regions with random environmental conditions. Furthermore, the proposed intelligent algorithms are more attractive as the power of the PV system to be used is high.